BioDT - The Recreational Model

1. Scope of the Model

The recreational potential model (RP Model) was developed as part of the Cultural Ecosystem Services prototype Digital Twin (CES-pDT) workpackage within the BioDT project. The (CES-pDT) has two independent core models:

• the Biodiversity model (developed by Simon Rolph and Dylan Carbone, UK Centre for Ecology & Hydrology, Wallingford, United Kingdom): aims at estimating biodiversity levels across mammals, birds, plant and insects.

• the Recreational Potential model (developed by Jan Dick, Chris Andrews, Maddalena Tigli, Megan Williams and Joe Marsh Rossney, UK Centre for Ecology & Hydrology, Edinburgh, United Kingdom): aims at estimating the landscapes’ capacity to provide opportunities for outdoor recreation based on varying user interests.

This report specifically documents the Recreational Potential (RP) Model, detailing its methodology, data sources, operational mechanisms, and outlining areas for future refinement and development.

CHRISTOPHER TO ADD MORE INFO? MAYBE THERE ARE SOME LIT SOURCES USED?

2. The model

2.1. Model description

flowchart LR
 subgraph s1["compute recreational potential"]
        C[("cropping of the 4<br>components file to the<br>extent of interest")]
        E["Compute each component's<br>contribution"]
        F["Rescale to unit interval"]
  end
    A("Definition of the<br><i>persona preferences</i>") ---> E
    B("Definition of the<br><i>area of interest</i>") --> C
    C --> E
    E --> F
    F --> G["recreational potential spatRast"]
    style s1 fill:#FFF9C4
    style A color:#000000
    style B color:#000000
    style C color:#000000
    style E color:#000000
    style F color:#000000
    style G color:#000000

The Recreational Potential (RP) Model estimates the recreational value of landscapes using four key components, each represented by a raster file:

1. Landscape Component (SLSRA.tif)

2. Natural Features Component (FIPS_N.tif)

3. Infrastructure Component (FIPS_I.tif)

4. Water Component (Water.tif)

Each of the four component raster files contains multiple layers, with each layer representing a specific feature. Each raster covers the entire domain of the model (with a 20mx20m resolution), currently Scotland (see 2.3. The underlying data).

The values in these raster layers range from 0 to 1, interpreted as follows:

• 1 indicates that the feature is present in that cell.

• 0 or NAs indicates that the feature is absent from that cell

• Values between 0 and 1 are present in the features/layers of the Infrastructure and Water components rasters, and reflect the proximity from nearby features. This continuous scale allows the RP model to incorporate how areas near e.g., a loch, can still contribute to the recreational potential, even if they do not directly contain the feature

The four rasters form the basis for calculating the Recreational Potential (RP) across the landscape.

To run the RP model the user must define:

• Area of interest: defined using a bounding box (terra::ext() in R).

• Persona preferences: a CSV file assigning a score (0–10) to each of the 87 features (see 2.3. The underlying data), reflecting the persona’s interest in each (see example persona file “presets.csv” ).

RP model processing steps

1. Cropping the raster files

The model first crops each of the four component rasters to the specified area of interest. Using the terra package in R, the bounding box is converted to a vector polygon with terra::vect(), and then used to crop the raster via terra::crop().

2. Calculating each components contribution

For each cell (20m x 20m), the model extracts values from all relevant layers (features), multiplies them by the corresponding persona score, and computes a weighted sum across layers. This results in one new raster per component that reflects the component’s contribution based on the persona preferences.

3. Re-scale to unit interval

Each of the components contribution rasters is re-scaled using “min-max normalization”:

\[ scaled.value = (x-min)/mix-min \]

Minimum values are mapped as 0, maximum values are mapped as 1, and all the intermediate values are proportionally scaled.

4. Combining the components RP

The four normalized component rasters are summed cell by cell. The resulting raster is then re-scaled again (min-max normalization) to produce the final Recreational Potential raster.

The model’s output is a SpatRaster object (that can be saved as a .tif) with five layers:

1. “SLSRA”: the RP of the landscape component

2. “FIPS_N”: the RP of the natural features component

3. “FIPS_I”: the RP of the infrastructure component

4. “Water”: the RP of the water component

5. “Recreational_Potential”: the combined RP

In all layers, values range from 0 to 1, where higher values (closer to 1) indicate greater recreational potential.

Because the model output is a result of the normalization of recreational potential of a specific area of interest, the values are relative within that area. This means that running the model with a different geographic extent will yield to different scores for the same features, even if the underlying data is unchanged.

2.2. Example run

We demonstrate the RP Model applied to the Bush Estate area (nearby Edinburgh) using two contrasting personas: a hard recreationalist and a soft recreationalist. The resulting maps illustrate how different preferences lead to different areas being highlighted for recreational value.

1. Hard recreationalist exmaple

Here we show a model run the Bush Estate area using a “hard recreationalist” example of persona (saved as “Hard_Recreationalist” in “presets.csv” ). Key features preferences for this persona are:

Features scored the highest and the lowest in the hard recreationalist persona

score_group	features
highest scores (scored 10 or 9)	Rock Walls (FIPS_N), Mountains (FIPS_N), Inland cliffs, rock pavements and outcrops (SLSRA), Rock cliffs, ledges and shores (SLSRA), National Park (SLSRA)
lowest scores (scored 0 or 1)	Built-up areas (FIPS_N), Flood plain (FIPS_N), Depressions (FIPS_N), No slope (FIPS_N), Gentle slope (FIPS_N), Country Park (SLSRA), Raised and blanket bog (SLSRA), Valley mires, poor fens and transition mires (SLSRA), Windthrown woodland (SLSRA), Woodland fringes and clearings and tall forb stands (SLSRA), Bare field or exposed soil (SLSRA), Built-up area (SLSRA), Royal Society for the Protection of Birds (RSPB) Reserve (SLSRA), Pond (Water), Motorway (FIPS_I), A Road (FIPS_I), B Road (FIPS_I), Minor or local road (FIPS_I), Access roads or Track (FIPS_I), Saltings (FIPS_N)

This persona highly values remote, challenging environments and avoids built-up or highly managed areas.

2. Soft recreationalist example

Here we show a model run the Bush Estate area using a “soft recreationalist” example of persona (saved as “Soft_Recreationalist” in “presets.csv”). Key features preferences for this persona are:

Features scored the highest and the lowest in the soft recreationalist persona

score_group	features
highest scores (scored 10 or 9)	Traffic Free: Paved Surface (FIPS_I), Coastal dunes and sandy shore (SLSRA), National Park (SLSRA), Major Lochs (Water), Beaches or Dunes (FIPS_N), No slope (FIPS_N), Freshwater (SLSRA), Mixed deciduous and coniferous woodland (SLSRA), Scots pine woodland (SLSRA), Broadleaved deciduous woodland (SLSRA)
lowest scores (scored 0 or 1)	Rock Walls (FIPS_N), Built-up areas (FIPS_N), Rocks or Scree (FIPS_N), Depressions (FIPS_N), Extremely steep slope (FIPS_N), Screes (SLSRA), Windthrown woodland (SLSRA), Bare field or exposed soil (SLSRA), Unnamed minor stream or tributary (Water), Motorway (FIPS_I), A Road (FIPS_I), Saltings (FIPS_N)

Soft recreationalists are drawn to tranquil, accessible landscapes and natural water features, avoiding steep or harsh terrain.

2.3. The underlying data

The methodology used to create each of the four components’ Scotland wide raster file is described in detail in this section.

An overview of the raster files of all four component for the Easter Bush area (the same extent displayed in the example model run) is also displayed.

Description of the 4 components’ raster files.

name raster	description
SLSRA.tif	Landscape component This includes data on land cover type, landscape designations and conservation, and farmland of high nature value. resolution: 20x20 extent: whole of Scotland crs: British National Grid nr. features (layers): 40
FIPS_N.tif	Natural Features component This includes data on land form types, soil types and slope. resolution: 20x20 extent: whole of Scotland crs: British National Grid nr. features (layers): 24
Water.tif	Water component This includes data on water feature types, as the presence of a lake or river. resolution: 20x20 extent: whole of Scotland crs: British National Grid nr. features (layers): 13
FIPS_I.tif	Infrastructure component This includes data on road and track, footpaths and cycle networks. resolution: 20x20 extent: whole of Scotland crs: British National Grid nr. features (layers): 10

1. Landscape component

CHRISTOPHER TO PROVIDE INFO ON THIS RASTRE FILE

Landscape component features

nr	Name	Description
1	SLSRA_CP_2	Country Park
2	SLSRA_HNV_2	Designated High Nature Value (HNV) farmland
3	SLSRA_LCM_1	Alpine and subalpine grassland
4	SLSRA_LCM_2	Arable land and market gardens
5	SLSRA_LCM_3	Arctic, alpine and subalpine scrub
6	SLSRA_LCM_4	Bare field or exposed soil
7	SLSRA_LCM_5	Base-rich fens and calcareous spring mires
8	SLSRA_LCM_6	Broadleaved deciduous woodland
9	SLSRA_LCM_7	Built-up area
10	SLSRA_LCM_8	Coastal dunes and sandy shore
11	SLSRA_LCM_9	Coastal shingle
12	SLSRA_LCM_10	Dry grassland
13	SLSRA_LCM_11	Freshwater
14	SLSRA_LCM_12	Inland cliffs, rock pavements and outcrops
15	SLSRA_LCM_13	Lines of trees, small planted woodlands, early-stage woodland and coppice
16	SLSRA_LCM_14	Littoral sediment or saltmarsh
17	SLSRA_LCM_15	Mesic grassland
18	SLSRA_LCM_16	Mixed deciduous and coniferous woodland
19	SLSRA_LCM_17	Non-native coniferous plantation
20	SLSRA_LCM_18	Raised and blanket bog
21	SLSRA_LCM_19	Riverine and fen scrubs
22	SLSRA_LCM_20	Rock cliffs, ledges and shores
23	SLSRA_LCM_21	Scots pine woodland
24	SLSRA_LCM_22	Screes
25	SLSRA_LCM_23	Seasonally wet and wet grassland
26	SLSRA_LCM_24	Temperate montane scrub
27	SLSRA_LCM_25	Temperate shrub heathland
28	SLSRA_LCM_26	Valley mires, poor fens and transition mires
29	SLSRA_LCM_27	Windthrown woodland
30	SLSRA_LCM_28	Woodland fringes and clearings and tall forb stands
31	SLSRA_NNR_2	National Nature Reserve (NNR)
32	SLSRA_NP_2	National Park
33	SLSRA_NR_2	Nature Reserve
34	SLSRA_RP_2	Regional Park
35	SLSRA_RSPB_2	Royal Society for the Protection of Birds (RSPB) Reserve
36	SLSRA_SAC_2	Special Area of Conservation (SAC)
37	SLSRA_SPA_2	Special Protection Area (SPA)
38	SLSRA_SSSI_2	Site of Special Scientific Interest (SSSI)
39	SLSRA_SWT_2	Scottish Wildlife Trust Reserve
40	SLSRA_WLA_2	Wild Land Areas

2. Natural Features component

CHRISTOPHER TO PROVIDE INFO ON THIS RASTRE FILE

Natural Features component features

nr	Name	Description
1	FIPS_N_Landform_1	Foothills
2	FIPS_N_Landform_2	Mountains
3	FIPS_N_Landform_3	Terraces
4	FIPS_N_Landform_4	Flood plain
5	FIPS_N_Landform_5	Beaches or Dunes
6	FIPS_N_Landform_6	Rocks or Scree
7	FIPS_N_Landform_7	Depressions
8	FIPS_N_Landform_8	Hills
9	FIPS_N_Landform_9	Lowlands
10	FIPS_N_Landform_10	Rock Walls
11	FIPS_N_Landform_11	Uplands
12	FIPS_N_Landform_12	Valley sides
13	FIPS_N_Landform_13	Valley bottom
14	FIPS_N_Landform_14	Built-up areas
15	FIPS_N_Landform_15	Saltings
16	FIPS_N_Landform_16	Hummocks, mounds or moraines
17	FIPS_N_Slope_1	No slope
18	FIPS_N_Slope_2	Gentle slope
19	FIPS_N_Slope_3	Medium slope
20	FIPS_N_Slope_4	Steep slope
21	FIPS_N_Slope_5	Very steep slope
22	FIPS_N_Slope_6	Extremely steep slope
23	FIPS_N_Soil_1	Peat or Organic
24	FIPS_N_Soil_2	Mineral

3. Infrastructure component

Infrastructure component features

nr	Name	Description
1	FIPS_I_LocalPathNetwork_2	Path
2	FIPS_I_RoadsTracks_4	Minor or local road
3	FIPS_I_RoadsTracks_2	A Road
4	FIPS_I_RoadsTracks_1	Motorway
5	FIPS_I_RoadsTracks_5	Access roads or Track
6	FIPS_I_RoadsTracks_3	B Road
7	FIPS_I_NationalCycleNetwork_1	On Road: Paved Surface
8	FIPS_I_NationalCycleNetwork_2	Traffic Free: Unpaved Surface
9	FIPS_I_NationalCycleNetwork_3	Traffic Free: Paved Surface
10	FIPS_I_NationalCycleNetwork_4	On Road: Unpaved Surface

The infrastructure component is derived from a raster indicating the presence or absence of infrastructure features. This original raster, labeled “original” in the Easter Bush example map, contains: 1 for cells where the feature is present, and 0 or NAs for cells where the feature is absent. CHRISTOPHER TO ADD WHERE THE ORIGINAL RASTER CAME FROM.

To account for proximity effects, the distance from each cell to the nearest infrastructure feature was calculated using the terra::distance() function in R. The resulting raster, labeled “distance” in the example map, contains the distance in meters from the center of each cell to the nearest feature.
Because this operation is memory-intensive at the national scale, the original Scotland-wide raster was divided into 20 spatial “windows”.To ensure accuracy near window boundaries, each window was processed with an additional 10 km buffer, allowing distances to be computed to features located in neighboring windows.

Finally, considering only the cells that were <= 500m distance from a feature (everything with features further away than that got assigned a NAs), the values were re-scaled from 0 to 1 (1 = present in cell, and then decreasing score as you get away from that cell). This raster is named “scored” in the example map, using the distance decay function JOE TO ADD PAPER FOR THIS FUNCTION + kappa and alpha values:

\[ distance (m) = \frac{\kappa + 1}{\kappa + \exp(\alpha x)} \]

Where kappa has a value of XX and alpha has a value of XX.

The “scored” raster is what it is utilized by the RP model.

4. Water component

The water component is derived from a raster indicating the presence or absence of infrastructure features. This original raster, labeled “original” in the Easter Bush example map, contains: 1 for cells where the feature is present, and 0 or NAs for cells where the feature is absent. CHRISTOPHER TO ADD WHERE THE ORIGINAL RASTER CAME FROM.

Then, using the same methodology applied to the infrastructure component’s raster the “distance” and ”scored” raster were created.

The “scored” raster is what it is utilized by the RP model.

Water component features

nr	Name	Description
1	Water_Lakes_1	Pond
2	Water_Lakes_2	Lochan
3	Water_Lakes_3	Small Lochs
4	Water_Lakes_4	Medium Lochs
5	Water_Lakes_5	Large Lochs
6	Water_Lakes_6	Major Lochs
7	Water_Rivers_1	Minor river or tributary
8	Water_Rivers_2	Unnamed minor stream or tributary
9	Water_Rivers_3	Major river or tributary
10	Water_Rivers_4	Named minor stream or tributary
11	Water_Rivers_5	Lake
12	Water_Rivers_6	Tidal river or estuary
13	Water_Rivers_7	Canal

3. The Shiny app for the “watches” spin off

JOE TO ADD SECTION, MAYBE PROVIDE LINK?

4. Compertamentalization

JOE TO ADD SECTION, I HAVE NO IDEA WHAT WAS DONE HERE

5. Running the model for the whole of Scotland

MADDA TO ADD SECTION ONCE MODEL HAS RUN

6. Lessons learned

1. Transitioning to terra::SpatRaster Improved Performance

One of the most significant improvements came from switching from the legacy raster::Raster to the more modern and efficient terra::SpatRaster format. This change brought several key benefits:

• Faster processing of raster operations such as cropping, masking, and mathematical transformations.

• Better memory management, particularly for large datasets covering all of Scotland with a high resolutions.

• Improved integration with spatial vector formats and compatibility with modern spatial workflows in R.

Overall, the use of the package terra reduced computational overhead significantly (see report “Optimizing BioDT Recreational Potential (RT) model run-time”).

2. Pre-processing at the National Scale Greatly Reduces Run-time

Pre-processing component rasters at the national (Scotland-wide) scale, rather than performing these operations at run-time for each user-defined area. Pre-processing included:

• Standardizing spatial properties (extent, resolution, projection) across all four component rasters to ensure seamless overlay and computation.

• Pre-computing distance rasters, to avoid repeated and costly distance calculations during model runs.

• Pre-calculating scored rasters using predefined thresholds and decay functions, making them immediately usable for user-defined queries.

Overall, the use of this approach has reduced computational overhead significantly (see report “Optimizing BioDT Recreational Potential (RT) model run-time”).

3. JOE to add lessons on shiny/compartmentalization if any
4. CHRIS to add any other lessons if he has any

6. Future improvements

As the RP Model was developed as a prototype, several enhancements could improve its usability, accuracy, and performance. The following areas have been identified for future development:

1. Simplifying feature scoring

   Currently, the model requires users to score 87 individual features across the four components. While this allows for a detailed definition of the “persona profile”, it can be time-consuming for users. A future version could:

   • Group similar features into broader thematic categories (e.g., “mountain terrain,” “urban infrastructure,” “wetlands”), allowing users to score these aggregated categories rather than individual layers, making the scoring process more intuitive and efficient

   • Maintain the option to score each of the 87 featre for expert users who want fine control

   A simpler model configuration could make the model accessible to more users.

2. Connecting to Live, regularly updated data-sets

   Currently the model relies on static raster layers to represents the components, which may become outdated as landscapes change or as better data becomes available. A future version of the model could:

   • Link the model to live/regularly updated data-sets
   • Automatically generate the “original” raster layers for each component when updated data becomes available

3. User-defined weighting components

   Currently the RP model assumes that all four components contribute equally to the Recreational Potential score. However, not all personas may value these components equally. A future version of the model could:

   • Allow users to specify custom weights for each of the four components (e.g. ,50% landscape, 20% water, 15% infrastructure and 15% natural features)

4. Expanding the number of Persona presets available

   To make the RP model more accessible to casual non-technical users, and to encourage exploration and experimentation with it, a future version of the model could:

   • Provide persona preset for common user types (e.g., hikers, birdwatchers, families, cyclists etc.)
   • Define both features scores and components weights for each persona

5. Include seasonal dynamics

   Currently, the RP model assumes a static Recreation Potential throughout the year, however, this can vary with seasons, and weather conditions. A future version of the model could:

   • integrate temporal layers by asking the persona to specify what season they are intending to recreate in. Then the model could have the underlying components starting values to have different values for different seasons, accounting for trail closures or flooding risks etc.)

6. Allow for “future scenarios”

   With the aim of making the RP model the more relevant to policy makers and stakeholders, a future version of the model could:

   • have a “scenario analysis” mode, where users can “add” or “remove” values underlying components (e.g., to simulate a new trail or a new cycle path being added or the change of a landscape due to climate change or socio-economic changes)
   • allow for direct comparison on how the two (or more) scenarios differ (provide the user with a decrease or increase of Recreational Potential raster.

7. JOE and CHRIS to add something if you have other ideas

References